Waverider: Leveraging Hierarchical, Multi-Resolution Maps for Efficient and Reactive Obstacle Avoidance

TL;DR

The paper tackles real-time obstacle avoidance for mobile robots with large perceptive radii under tight compute budgets. It proposes a multi-resolution reactive navigation system that leverages wavemap hierarchical maps and Riemannian Motion Policies to generate and fuse obstacle-avoidance policies across scales, enabling high-rate operation without expensive pre-processing. Key contributions include an efficient hierarchical policy generation method, a numerical analysis of hierarchical policy approximation errors, and extensive simulations and real MAV experiments showing favorable runtime, safety, and robustness compared to CHOMP. This approach enables safe, scalable navigation in 3D environments and ships as open-source, suitable for integration with additional objectives such as goal seeking or manipulation tasks.

Abstract

Fast and reliable obstacle avoidance is an important task for mobile robots. In this work, we propose an efficient reactive system that provides high-quality obstacle avoidance while running at hundreds of hertz with minimal resource usage. Our approach combines wavemap, a hierarchical volumetric map representation, with a novel hierarchical and parallelizable obstacle avoidance algorithm formulated through Riemannian Motion Policies (RMP). Leveraging multi-resolution obstacle avoidance policies, the proposed navigation system facilitates precise, low-latency (36ms), and extremely efficient obstacle avoidance with a very large perceptive radius (30m). We perform extensive statistical evaluations on indoor and outdoor maps, verifying that the proposed system compares favorably to fixed-resolution RMP variants and CHOMP. Finally, the RMP formulation allows the seamless fusion of obstacle avoidance with additional objectives, such as goal-seeking, to obtain a fully-fledged navigation system that is versatile and robust. We deploy the system on a Micro Aerial Vehicle and show how it navigates through an indoor obstacle course. Our complete implementation, called waverider, is made available as open source.

Figures (12)

• Figure 1: Example trajectories comparing our multi-resolution collision avoidance method (red) to equivalent RMP-based formulations that consider all obstacles at the highest resolution within a radius of 1m (green) and 3m (blue). The fixed-resolution RMP trajectories are jerkier and more prone to get stuck (top-left). CHOMP (brown) yields smooth, albeit overly cautious trajectories and occasionally cuts through obstacles (top-right, bottom-left).
• Figure 2: Block diagram of the proposed navigation system. External components are highlighted in yellow, tightly integrated components in blue, and new components introduced in this paper in green.
• Figure 3: Comparison of an environment represented using fixed-resolution (left) and hierarchical obstacle cells (right). Our approach uses hierarchical cells, whose resolution (light brown to dark green) is high close to the robot (red) and decreases with distance.
• Figure 4: Left: Perceptive radius defined by $d_{max}( \lambda)$ as used in the obstacle filter (red). Limited, fixed-resolution comparison variants used in \ref{['sec:statistic_eval']} are marked with a blue resp. green cross. Right: Worst-case counts of voxels to visit. Even with small perceptive radii, the fixed-resolution variants need to potentially iterate over significantly more voxels to provide the same quality of obstacle avoidance (log-scale).
• Figure 5: Example of obstacles that can be either modeled by a single, large policy ($P_{F}$) or multiple small, high-resolution policies ($P^i_f$). The distance $d$ represents the distance from the robot to the center of the obstacle block.